2 * Copyright (c) 2012 Clément Bœsch
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * EBU R.128 implementation
24 * @see http://tech.ebu.ch/loudness
25 * @see https://www.youtube.com/watch?v=iuEtQqC-Sqo "EBU R128 Introduction - Florian Camerer"
26 * @todo implement start/stop/reset through filter command injection
27 * @todo support other frequencies to avoid resampling
32 #include "libavutil/avassert.h"
33 #include "libavutil/avstring.h"
34 #include "libavutil/channel_layout.h"
35 #include "libavutil/dict.h"
36 #include "libavutil/ffmath.h"
37 #include "libavutil/xga_font_data.h"
38 #include "libavutil/opt.h"
39 #include "libavutil/timestamp.h"
40 #include "libswresample/swresample.h"
46 #define MAX_CHANNELS 63
48 /* pre-filter coefficients */
49 #define PRE_B0 1.53512485958697
50 #define PRE_B1 -2.69169618940638
51 #define PRE_B2 1.19839281085285
52 #define PRE_A1 -1.69065929318241
53 #define PRE_A2 0.73248077421585
55 /* RLB-filter coefficients */
59 #define RLB_A1 -1.99004745483398
60 #define RLB_A2 0.99007225036621
62 #define ABS_THRES -70 ///< silence gate: we discard anything below this absolute (LUFS) threshold
63 #define ABS_UP_THRES 10 ///< upper loud limit to consider (ABS_THRES being the minimum)
64 #define HIST_GRAIN 100 ///< defines histogram precision
65 #define HIST_SIZE ((ABS_UP_THRES - ABS_THRES) * HIST_GRAIN + 1)
68 * A histogram is an array of HIST_SIZE hist_entry storing all the energies
69 * recorded (with an accuracy of 1/HIST_GRAIN) of the loudnesses from ABS_THRES
70 * (at 0) to ABS_UP_THRES (at HIST_SIZE-1).
71 * This fixed-size system avoids the need of a list of energies growing
72 * infinitely over the time and is thus more scalable.
75 int count; ///< how many times the corresponding value occurred
76 double energy; ///< E = 10^((L + 0.691) / 10)
77 double loudness; ///< L = -0.691 + 10 * log10(E)
81 double *cache[MAX_CHANNELS]; ///< window of filtered samples (N ms)
82 int cache_pos; ///< focus on the last added bin in the cache array
83 double sum[MAX_CHANNELS]; ///< sum of the last N ms filtered samples (cache content)
84 int filled; ///< 1 if the cache is completely filled, 0 otherwise
85 double rel_threshold; ///< relative threshold
86 double sum_kept_powers; ///< sum of the powers (weighted sums) above absolute threshold
87 int nb_kept_powers; ///< number of sum above absolute threshold
88 struct hist_entry *histogram; ///< histogram of the powers, used to compute LRA and I
91 struct rect { int x, y, w, h; };
93 typedef struct EBUR128Context {
94 const AVClass *class; ///< AVClass context for log and options purpose
97 int peak_mode; ///< enabled peak modes
98 double *true_peaks; ///< true peaks per channel
99 double *sample_peaks; ///< sample peaks per channel
100 double *true_peaks_per_frame; ///< true peaks in a frame per channel
101 #if CONFIG_SWRESAMPLE
102 SwrContext *swr_ctx; ///< over-sampling context for true peak metering
103 double *swr_buf; ///< resampled audio data for true peak metering
108 int do_video; ///< 1 if video output enabled, 0 otherwise
109 int w, h; ///< size of the video output
110 struct rect text; ///< rectangle for the LU legend on the left
111 struct rect graph; ///< rectangle for the main graph in the center
112 struct rect gauge; ///< rectangle for the gauge on the right
113 AVFrame *outpicref; ///< output picture reference, updated regularly
114 int meter; ///< select a EBU mode between +9 and +18
115 int scale_range; ///< the range of LU values according to the meter
116 int y_zero_lu; ///< the y value (pixel position) for 0 LU
117 int y_opt_max; ///< the y value (pixel position) for 1 LU
118 int y_opt_min; ///< the y value (pixel position) for -1 LU
119 int *y_line_ref; ///< y reference values for drawing the LU lines in the graph and the gauge
122 int nb_channels; ///< number of channels in the input
123 double *ch_weighting; ///< channel weighting mapping
124 int sample_count; ///< sample count used for refresh frequency, reset at refresh
127 * The mult by 3 in the following is for X[i], X[i-1] and X[i-2] */
128 double x[MAX_CHANNELS * 3]; ///< 3 input samples cache for each channel
129 double y[MAX_CHANNELS * 3]; ///< 3 pre-filter samples cache for each channel
130 double z[MAX_CHANNELS * 3]; ///< 3 RLB-filter samples cache for each channel
132 #define I400_BINS (48000 * 4 / 10)
133 #define I3000_BINS (48000 * 3)
134 struct integrator i400; ///< 400ms integrator, used for Momentary loudness (M), and Integrated loudness (I)
135 struct integrator i3000; ///< 3s integrator, used for Short term loudness (S), and Loudness Range (LRA)
137 /* I and LRA specific */
138 double integrated_loudness; ///< integrated loudness in LUFS (I)
139 double loudness_range; ///< loudness range in LU (LRA)
140 double lra_low, lra_high; ///< low and high LRA values
143 int loglevel; ///< log level for frame logging
144 int metadata; ///< whether or not to inject loudness results in frames
145 int dual_mono; ///< whether or not to treat single channel input files as dual-mono
146 double pan_law; ///< pan law value used to calculate dual-mono measurements
147 int target; ///< target level in LUFS used to set relative zero LU in visualization
148 int gauge_type; ///< whether gauge shows momentary or short
149 int scale; ///< display scale type of statistics
154 PEAK_MODE_SAMPLES_PEAKS = 1<<1,
155 PEAK_MODE_TRUE_PEAKS = 1<<2,
159 GAUGE_TYPE_MOMENTARY = 0,
160 GAUGE_TYPE_SHORTTERM = 1,
164 SCALE_TYPE_ABSOLUTE = 0,
165 SCALE_TYPE_RELATIVE = 1,
168 #define OFFSET(x) offsetof(EBUR128Context, x)
169 #define A AV_OPT_FLAG_AUDIO_PARAM
170 #define V AV_OPT_FLAG_VIDEO_PARAM
171 #define F AV_OPT_FLAG_FILTERING_PARAM
172 static const AVOption ebur128_options[] = {
173 { "video", "set video output", OFFSET(do_video), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, V|F },
174 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "640x480"}, 0, 0, V|F },
175 { "meter", "set scale meter (+9 to +18)", OFFSET(meter), AV_OPT_TYPE_INT, {.i64 = 9}, 9, 18, V|F },
176 { "framelog", "force frame logging level", OFFSET(loglevel), AV_OPT_TYPE_INT, {.i64 = -1}, INT_MIN, INT_MAX, A|V|F, "level" },
177 { "info", "information logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_INFO}, INT_MIN, INT_MAX, A|V|F, "level" },
178 { "verbose", "verbose logging level", 0, AV_OPT_TYPE_CONST, {.i64 = AV_LOG_VERBOSE}, INT_MIN, INT_MAX, A|V|F, "level" },
179 { "metadata", "inject metadata in the filtergraph", OFFSET(metadata), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|V|F },
180 { "peak", "set peak mode", OFFSET(peak_mode), AV_OPT_TYPE_FLAGS, {.i64 = PEAK_MODE_NONE}, 0, INT_MAX, A|F, "mode" },
181 { "none", "disable any peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_NONE}, INT_MIN, INT_MAX, A|F, "mode" },
182 { "sample", "enable peak-sample mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_SAMPLES_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
183 { "true", "enable true-peak mode", 0, AV_OPT_TYPE_CONST, {.i64 = PEAK_MODE_TRUE_PEAKS}, INT_MIN, INT_MAX, A|F, "mode" },
184 { "dualmono", "treat mono input files as dual-mono", OFFSET(dual_mono), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, A|F },
185 { "panlaw", "set a specific pan law for dual-mono files", OFFSET(pan_law), AV_OPT_TYPE_DOUBLE, {.dbl = -3.01029995663978}, -10.0, 0.0, A|F },
186 { "target", "set a specific target level in LUFS (-23 to 0)", OFFSET(target), AV_OPT_TYPE_INT, {.i64 = -23}, -23, 0, V|F },
187 { "gauge", "set gauge display type", OFFSET(gauge_type), AV_OPT_TYPE_INT, {.i64 = 0 }, GAUGE_TYPE_MOMENTARY, GAUGE_TYPE_SHORTTERM, V|F, "gaugetype" },
188 { "momentary", "display momentary value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_MOMENTARY}, INT_MIN, INT_MAX, V|F, "gaugetype" },
189 { "m", "display momentary value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_MOMENTARY}, INT_MIN, INT_MAX, V|F, "gaugetype" },
190 { "shortterm", "display short-term value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_SHORTTERM}, INT_MIN, INT_MAX, V|F, "gaugetype" },
191 { "s", "display short-term value", 0, AV_OPT_TYPE_CONST, {.i64 = GAUGE_TYPE_SHORTTERM}, INT_MIN, INT_MAX, V|F, "gaugetype" },
192 { "scale", "sets display method for the stats", OFFSET(scale), AV_OPT_TYPE_INT, {.i64 = 0}, SCALE_TYPE_ABSOLUTE, SCALE_TYPE_RELATIVE, V|F, "scaletype" },
193 { "absolute", "display absolute values (LUFS)", 0, AV_OPT_TYPE_CONST, {.i64 = SCALE_TYPE_ABSOLUTE}, INT_MIN, INT_MAX, V|F, "scaletype" },
194 { "LUFS", "display absolute values (LUFS)", 0, AV_OPT_TYPE_CONST, {.i64 = SCALE_TYPE_ABSOLUTE}, INT_MIN, INT_MAX, V|F, "scaletype" },
195 { "relative", "display values relative to target (LU)", 0, AV_OPT_TYPE_CONST, {.i64 = SCALE_TYPE_RELATIVE}, INT_MIN, INT_MAX, V|F, "scaletype" },
196 { "LU", "display values relative to target (LU)", 0, AV_OPT_TYPE_CONST, {.i64 = SCALE_TYPE_RELATIVE}, INT_MIN, INT_MAX, V|F, "scaletype" },
200 AVFILTER_DEFINE_CLASS(ebur128);
202 static const uint8_t graph_colors[] = {
203 0xdd, 0x66, 0x66, // value above 1LU non reached below -1LU (impossible)
204 0x66, 0x66, 0xdd, // value below 1LU non reached below -1LU
205 0x96, 0x33, 0x33, // value above 1LU reached below -1LU (impossible)
206 0x33, 0x33, 0x96, // value below 1LU reached below -1LU
207 0xdd, 0x96, 0x96, // value above 1LU line non reached below -1LU (impossible)
208 0x96, 0x96, 0xdd, // value below 1LU line non reached below -1LU
209 0xdd, 0x33, 0x33, // value above 1LU line reached below -1LU (impossible)
210 0x33, 0x33, 0xdd, // value below 1LU line reached below -1LU
211 0xdd, 0x66, 0x66, // value above 1LU non reached above -1LU
212 0x66, 0xdd, 0x66, // value below 1LU non reached above -1LU
213 0x96, 0x33, 0x33, // value above 1LU reached above -1LU
214 0x33, 0x96, 0x33, // value below 1LU reached above -1LU
215 0xdd, 0x96, 0x96, // value above 1LU line non reached above -1LU
216 0x96, 0xdd, 0x96, // value below 1LU line non reached above -1LU
217 0xdd, 0x33, 0x33, // value above 1LU line reached above -1LU
218 0x33, 0xdd, 0x33, // value below 1LU line reached above -1LU
221 static const uint8_t *get_graph_color(const EBUR128Context *ebur128, int v, int y)
223 const int above_opt_max = y > ebur128->y_opt_max;
224 const int below_opt_min = y < ebur128->y_opt_min;
225 const int reached = y >= v;
226 const int line = ebur128->y_line_ref[y] || y == ebur128->y_zero_lu;
227 const int colorid = 8*below_opt_min+ 4*line + 2*reached + above_opt_max;
228 return graph_colors + 3*colorid;
231 static inline int lu_to_y(const EBUR128Context *ebur128, double v)
233 v += 2 * ebur128->meter; // make it in range [0;...]
234 v = av_clipf(v, 0, ebur128->scale_range); // make sure it's in the graph scale
235 v = ebur128->scale_range - v; // invert value (y=0 is on top)
236 return v * ebur128->graph.h / ebur128->scale_range; // rescale from scale range to px height
242 static const uint8_t font_colors[] = {
247 static void drawtext(AVFrame *pic, int x, int y, int ftid, const uint8_t *color, const char *fmt, ...)
255 if (ftid == FONT16) font = avpriv_vga16_font, font_height = 16;
256 else if (ftid == FONT8) font = avpriv_cga_font, font_height = 8;
260 vsnprintf(buf, sizeof(buf), fmt, vl);
263 for (i = 0; buf[i]; i++) {
265 uint8_t *p = pic->data[0] + y*pic->linesize[0] + (x + i*8)*3;
267 for (char_y = 0; char_y < font_height; char_y++) {
268 for (mask = 0x80; mask; mask >>= 1) {
269 if (font[buf[i] * font_height + char_y] & mask)
272 memcpy(p, "\x00\x00\x00", 3);
275 p += pic->linesize[0] - 8*3;
280 static void drawline(AVFrame *pic, int x, int y, int len, int step)
283 uint8_t *p = pic->data[0] + y*pic->linesize[0] + x*3;
285 for (i = 0; i < len; i++) {
286 memcpy(p, "\x00\xff\x00", 3);
291 static int config_video_output(AVFilterLink *outlink)
295 AVFilterContext *ctx = outlink->src;
296 EBUR128Context *ebur128 = ctx->priv;
299 /* check if there is enough space to represent everything decently */
300 if (ebur128->w < 640 || ebur128->h < 480) {
301 av_log(ctx, AV_LOG_ERROR, "Video size %dx%d is too small, "
302 "minimum size is 640x480\n", ebur128->w, ebur128->h);
303 return AVERROR(EINVAL);
305 outlink->w = ebur128->w;
306 outlink->h = ebur128->h;
307 outlink->sample_aspect_ratio = (AVRational){1,1};
311 /* configure text area position and size */
312 ebur128->text.x = PAD;
313 ebur128->text.y = 40;
314 ebur128->text.w = 3 * 8; // 3 characters
315 ebur128->text.h = ebur128->h - PAD - ebur128->text.y;
317 /* configure gauge position and size */
318 ebur128->gauge.w = 20;
319 ebur128->gauge.h = ebur128->text.h;
320 ebur128->gauge.x = ebur128->w - PAD - ebur128->gauge.w;
321 ebur128->gauge.y = ebur128->text.y;
323 /* configure graph position and size */
324 ebur128->graph.x = ebur128->text.x + ebur128->text.w + PAD;
325 ebur128->graph.y = ebur128->gauge.y;
326 ebur128->graph.w = ebur128->gauge.x - ebur128->graph.x - PAD;
327 ebur128->graph.h = ebur128->gauge.h;
329 /* graph and gauge share the LU-to-pixel code */
330 av_assert0(ebur128->graph.h == ebur128->gauge.h);
332 /* prepare the initial picref buffer */
333 av_frame_free(&ebur128->outpicref);
334 ebur128->outpicref = outpicref =
335 ff_get_video_buffer(outlink, outlink->w, outlink->h);
337 return AVERROR(ENOMEM);
338 outpicref->sample_aspect_ratio = (AVRational){1,1};
340 /* init y references values (to draw LU lines) */
341 ebur128->y_line_ref = av_calloc(ebur128->graph.h + 1, sizeof(*ebur128->y_line_ref));
342 if (!ebur128->y_line_ref)
343 return AVERROR(ENOMEM);
345 /* black background */
346 memset(outpicref->data[0], 0, ebur128->h * outpicref->linesize[0]);
348 /* draw LU legends */
349 drawtext(outpicref, PAD, PAD+16, FONT8, font_colors+3, " LU");
350 for (i = ebur128->meter; i >= -ebur128->meter * 2; i--) {
351 y = lu_to_y(ebur128, i);
352 x = PAD + (i < 10 && i > -10) * 8;
353 ebur128->y_line_ref[y] = i;
354 y -= 4; // -4 to center vertically
355 drawtext(outpicref, x, y + ebur128->graph.y, FONT8, font_colors+3,
356 "%c%d", i < 0 ? '-' : i > 0 ? '+' : ' ', FFABS(i));
360 ebur128->y_zero_lu = lu_to_y(ebur128, 0);
361 ebur128->y_opt_max = lu_to_y(ebur128, 1);
362 ebur128->y_opt_min = lu_to_y(ebur128, -1);
363 p = outpicref->data[0] + ebur128->graph.y * outpicref->linesize[0]
364 + ebur128->graph.x * 3;
365 for (y = 0; y < ebur128->graph.h; y++) {
366 const uint8_t *c = get_graph_color(ebur128, INT_MAX, y);
368 for (x = 0; x < ebur128->graph.w; x++)
369 memcpy(p + x*3, c, 3);
370 p += outpicref->linesize[0];
373 /* draw fancy rectangles around the graph and the gauge */
374 #define DRAW_RECT(r) do { \
375 drawline(outpicref, r.x, r.y - 1, r.w, 3); \
376 drawline(outpicref, r.x, r.y + r.h, r.w, 3); \
377 drawline(outpicref, r.x - 1, r.y, r.h, outpicref->linesize[0]); \
378 drawline(outpicref, r.x + r.w, r.y, r.h, outpicref->linesize[0]); \
380 DRAW_RECT(ebur128->graph);
381 DRAW_RECT(ebur128->gauge);
386 static int config_audio_input(AVFilterLink *inlink)
388 AVFilterContext *ctx = inlink->dst;
389 EBUR128Context *ebur128 = ctx->priv;
391 /* Force 100ms framing in case of metadata injection: the frames must have
392 * a granularity of the window overlap to be accurately exploited.
393 * As for the true peaks mode, it just simplifies the resampling buffer
394 * allocation and the lookup in it (since sample buffers differ in size, it
395 * can be more complex to integrate in the one-sample loop of
396 * filter_frame()). */
397 if (ebur128->metadata || (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS))
398 inlink->min_samples =
399 inlink->max_samples =
400 inlink->partial_buf_size = inlink->sample_rate / 10;
404 static int config_audio_output(AVFilterLink *outlink)
407 AVFilterContext *ctx = outlink->src;
408 EBUR128Context *ebur128 = ctx->priv;
409 const int nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
411 #define BACK_MASK (AV_CH_BACK_LEFT |AV_CH_BACK_CENTER |AV_CH_BACK_RIGHT| \
412 AV_CH_TOP_BACK_LEFT|AV_CH_TOP_BACK_CENTER|AV_CH_TOP_BACK_RIGHT| \
413 AV_CH_SIDE_LEFT |AV_CH_SIDE_RIGHT| \
414 AV_CH_SURROUND_DIRECT_LEFT |AV_CH_SURROUND_DIRECT_RIGHT)
416 ebur128->nb_channels = nb_channels;
417 ebur128->ch_weighting = av_calloc(nb_channels, sizeof(*ebur128->ch_weighting));
418 if (!ebur128->ch_weighting)
419 return AVERROR(ENOMEM);
421 for (i = 0; i < nb_channels; i++) {
422 /* channel weighting */
423 const uint64_t chl = av_channel_layout_extract_channel(outlink->channel_layout, i);
424 if (chl & (AV_CH_LOW_FREQUENCY|AV_CH_LOW_FREQUENCY_2)) {
425 ebur128->ch_weighting[i] = 0;
426 } else if (chl & BACK_MASK) {
427 ebur128->ch_weighting[i] = 1.41;
429 ebur128->ch_weighting[i] = 1.0;
432 if (!ebur128->ch_weighting[i])
435 /* bins buffer for the two integration window (400ms and 3s) */
436 ebur128->i400.cache[i] = av_calloc(I400_BINS, sizeof(*ebur128->i400.cache[0]));
437 ebur128->i3000.cache[i] = av_calloc(I3000_BINS, sizeof(*ebur128->i3000.cache[0]));
438 if (!ebur128->i400.cache[i] || !ebur128->i3000.cache[i])
439 return AVERROR(ENOMEM);
442 #if CONFIG_SWRESAMPLE
443 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
446 ebur128->swr_buf = av_malloc_array(nb_channels, 19200 * sizeof(double));
447 ebur128->true_peaks = av_calloc(nb_channels, sizeof(*ebur128->true_peaks));
448 ebur128->true_peaks_per_frame = av_calloc(nb_channels, sizeof(*ebur128->true_peaks_per_frame));
449 ebur128->swr_ctx = swr_alloc();
450 if (!ebur128->swr_buf || !ebur128->true_peaks ||
451 !ebur128->true_peaks_per_frame || !ebur128->swr_ctx)
452 return AVERROR(ENOMEM);
454 av_opt_set_int(ebur128->swr_ctx, "in_channel_layout", outlink->channel_layout, 0);
455 av_opt_set_int(ebur128->swr_ctx, "in_sample_rate", outlink->sample_rate, 0);
456 av_opt_set_sample_fmt(ebur128->swr_ctx, "in_sample_fmt", outlink->format, 0);
458 av_opt_set_int(ebur128->swr_ctx, "out_channel_layout", outlink->channel_layout, 0);
459 av_opt_set_int(ebur128->swr_ctx, "out_sample_rate", 192000, 0);
460 av_opt_set_sample_fmt(ebur128->swr_ctx, "out_sample_fmt", outlink->format, 0);
462 ret = swr_init(ebur128->swr_ctx);
468 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS) {
469 ebur128->sample_peaks = av_calloc(nb_channels, sizeof(*ebur128->sample_peaks));
470 if (!ebur128->sample_peaks)
471 return AVERROR(ENOMEM);
477 #define ENERGY(loudness) (ff_exp10(((loudness) + 0.691) / 10.))
478 #define LOUDNESS(energy) (-0.691 + 10 * log10(energy))
479 #define DBFS(energy) (20 * log10(energy))
481 static struct hist_entry *get_histogram(void)
484 struct hist_entry *h = av_calloc(HIST_SIZE, sizeof(*h));
488 for (i = 0; i < HIST_SIZE; i++) {
489 h[i].loudness = i / (double)HIST_GRAIN + ABS_THRES;
490 h[i].energy = ENERGY(h[i].loudness);
495 static av_cold int init(AVFilterContext *ctx)
497 EBUR128Context *ebur128 = ctx->priv;
501 if (ebur128->loglevel != AV_LOG_INFO &&
502 ebur128->loglevel != AV_LOG_VERBOSE) {
503 if (ebur128->do_video || ebur128->metadata)
504 ebur128->loglevel = AV_LOG_VERBOSE;
506 ebur128->loglevel = AV_LOG_INFO;
509 if (!CONFIG_SWRESAMPLE && (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS)) {
510 av_log(ctx, AV_LOG_ERROR,
511 "True-peak mode requires libswresample to be performed\n");
512 return AVERROR(EINVAL);
515 // if meter is +9 scale, scale range is from -18 LU to +9 LU (or 3*9)
516 // if meter is +18 scale, scale range is from -36 LU to +18 LU (or 3*18)
517 ebur128->scale_range = 3 * ebur128->meter;
519 ebur128->i400.histogram = get_histogram();
520 ebur128->i3000.histogram = get_histogram();
521 if (!ebur128->i400.histogram || !ebur128->i3000.histogram)
522 return AVERROR(ENOMEM);
524 ebur128->integrated_loudness = ABS_THRES;
525 ebur128->loudness_range = 0;
527 /* insert output pads */
528 if (ebur128->do_video) {
531 .type = AVMEDIA_TYPE_VIDEO,
532 .config_props = config_video_output,
534 ret = ff_insert_outpad(ctx, 0, &pad);
539 .name = ebur128->do_video ? "out1" : "out0",
540 .type = AVMEDIA_TYPE_AUDIO,
541 .config_props = config_audio_output,
543 ret = ff_insert_outpad(ctx, ebur128->do_video, &pad);
548 av_log(ctx, AV_LOG_VERBOSE, "EBU +%d scale\n", ebur128->meter);
553 #define HIST_POS(power) (int)(((power) - ABS_THRES) * HIST_GRAIN)
555 /* loudness and power should be set such as loudness = -0.691 +
556 * 10*log10(power), we just avoid doing that calculus two times */
557 static int gate_update(struct integrator *integ, double power,
558 double loudness, int gate_thres)
561 double relative_threshold;
564 /* update powers histograms by incrementing current power count */
565 ipower = av_clip(HIST_POS(loudness), 0, HIST_SIZE - 1);
566 integ->histogram[ipower].count++;
568 /* compute relative threshold and get its position in the histogram */
569 integ->sum_kept_powers += power;
570 integ->nb_kept_powers++;
571 relative_threshold = integ->sum_kept_powers / integ->nb_kept_powers;
572 if (!relative_threshold)
573 relative_threshold = 1e-12;
574 integ->rel_threshold = LOUDNESS(relative_threshold) + gate_thres;
575 gate_hist_pos = av_clip(HIST_POS(integ->rel_threshold), 0, HIST_SIZE - 1);
577 return gate_hist_pos;
580 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
582 int i, ch, idx_insample;
583 AVFilterContext *ctx = inlink->dst;
584 EBUR128Context *ebur128 = ctx->priv;
585 const int nb_channels = ebur128->nb_channels;
586 const int nb_samples = insamples->nb_samples;
587 const double *samples = (double *)insamples->data[0];
588 AVFrame *pic = ebur128->outpicref;
590 #if CONFIG_SWRESAMPLE
591 if (ebur128->peak_mode & PEAK_MODE_TRUE_PEAKS) {
592 const double *swr_samples = ebur128->swr_buf;
593 int ret = swr_convert(ebur128->swr_ctx, (uint8_t**)&ebur128->swr_buf, 19200,
594 (const uint8_t **)insamples->data, nb_samples);
597 for (ch = 0; ch < nb_channels; ch++)
598 ebur128->true_peaks_per_frame[ch] = 0.0;
599 for (idx_insample = 0; idx_insample < ret; idx_insample++) {
600 for (ch = 0; ch < nb_channels; ch++) {
601 ebur128->true_peaks[ch] = FFMAX(ebur128->true_peaks[ch], fabs(*swr_samples));
602 ebur128->true_peaks_per_frame[ch] = FFMAX(ebur128->true_peaks_per_frame[ch],
610 for (idx_insample = 0; idx_insample < nb_samples; idx_insample++) {
611 const int bin_id_400 = ebur128->i400.cache_pos;
612 const int bin_id_3000 = ebur128->i3000.cache_pos;
614 #define MOVE_TO_NEXT_CACHED_ENTRY(time) do { \
615 ebur128->i##time.cache_pos++; \
616 if (ebur128->i##time.cache_pos == I##time##_BINS) { \
617 ebur128->i##time.filled = 1; \
618 ebur128->i##time.cache_pos = 0; \
622 MOVE_TO_NEXT_CACHED_ENTRY(400);
623 MOVE_TO_NEXT_CACHED_ENTRY(3000);
625 for (ch = 0; ch < nb_channels; ch++) {
628 if (ebur128->peak_mode & PEAK_MODE_SAMPLES_PEAKS)
629 ebur128->sample_peaks[ch] = FFMAX(ebur128->sample_peaks[ch], fabs(*samples));
631 ebur128->x[ch * 3] = *samples++; // set X[i]
633 if (!ebur128->ch_weighting[ch])
636 /* Y[i] = X[i]*b0 + X[i-1]*b1 + X[i-2]*b2 - Y[i-1]*a1 - Y[i-2]*a2 */
637 #define FILTER(Y, X, name) do { \
638 double *dst = ebur128->Y + ch*3; \
639 double *src = ebur128->X + ch*3; \
642 dst[0] = src[0]*name##_B0 + src[1]*name##_B1 + src[2]*name##_B2 \
643 - dst[1]*name##_A1 - dst[2]*name##_A2; \
646 // TODO: merge both filters in one?
647 FILTER(y, x, PRE); // apply pre-filter
648 ebur128->x[ch * 3 + 2] = ebur128->x[ch * 3 + 1];
649 ebur128->x[ch * 3 + 1] = ebur128->x[ch * 3 ];
650 FILTER(z, y, RLB); // apply RLB-filter
652 bin = ebur128->z[ch * 3] * ebur128->z[ch * 3];
654 /* add the new value, and limit the sum to the cache size (400ms or 3s)
655 * by removing the oldest one */
656 ebur128->i400.sum [ch] = ebur128->i400.sum [ch] + bin - ebur128->i400.cache [ch][bin_id_400];
657 ebur128->i3000.sum[ch] = ebur128->i3000.sum[ch] + bin - ebur128->i3000.cache[ch][bin_id_3000];
659 /* override old cache entry with the new value */
660 ebur128->i400.cache [ch][bin_id_400 ] = bin;
661 ebur128->i3000.cache[ch][bin_id_3000] = bin;
664 /* For integrated loudness, gating blocks are 400ms long with 75%
665 * overlap (see BS.1770-2 p5), so a re-computation is needed each 100ms
666 * (4800 samples at 48kHz). */
667 if (++ebur128->sample_count == 4800) {
668 double loudness_400, loudness_3000;
669 double power_400 = 1e-12, power_3000 = 1e-12;
670 AVFilterLink *outlink = ctx->outputs[0];
671 const int64_t pts = insamples->pts +
672 av_rescale_q(idx_insample, (AVRational){ 1, inlink->sample_rate },
675 ebur128->sample_count = 0;
677 #define COMPUTE_LOUDNESS(m, time) do { \
678 if (ebur128->i##time.filled) { \
679 /* weighting sum of the last <time> ms */ \
680 for (ch = 0; ch < nb_channels; ch++) \
681 power_##time += ebur128->ch_weighting[ch] * ebur128->i##time.sum[ch]; \
682 power_##time /= I##time##_BINS; \
684 loudness_##time = LOUDNESS(power_##time); \
687 COMPUTE_LOUDNESS(M, 400);
688 COMPUTE_LOUDNESS(S, 3000);
690 /* Integrated loudness */
691 #define I_GATE_THRES -10 // initially defined to -8 LU in the first EBU standard
693 if (loudness_400 >= ABS_THRES) {
694 double integrated_sum = 0;
695 int nb_integrated = 0;
696 int gate_hist_pos = gate_update(&ebur128->i400, power_400,
697 loudness_400, I_GATE_THRES);
699 /* compute integrated loudness by summing the histogram values
700 * above the relative threshold */
701 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
702 const int nb_v = ebur128->i400.histogram[i].count;
703 nb_integrated += nb_v;
704 integrated_sum += nb_v * ebur128->i400.histogram[i].energy;
707 ebur128->integrated_loudness = LOUDNESS(integrated_sum / nb_integrated);
708 /* dual-mono correction */
709 if (nb_channels == 1 && ebur128->dual_mono) {
710 ebur128->integrated_loudness -= ebur128->pan_law;
716 #define LRA_GATE_THRES -20
717 #define LRA_LOWER_PRC 10
718 #define LRA_HIGHER_PRC 95
720 /* XXX: example code in EBU 3342 is ">=" but formula in BS.1770
722 if (loudness_3000 >= ABS_THRES) {
724 int gate_hist_pos = gate_update(&ebur128->i3000, power_3000,
725 loudness_3000, LRA_GATE_THRES);
727 for (i = gate_hist_pos; i < HIST_SIZE; i++)
728 nb_powers += ebur128->i3000.histogram[i].count;
732 /* get lower loudness to consider */
734 nb_pow = LRA_LOWER_PRC * nb_powers / 100. + 0.5;
735 for (i = gate_hist_pos; i < HIST_SIZE; i++) {
736 n += ebur128->i3000.histogram[i].count;
738 ebur128->lra_low = ebur128->i3000.histogram[i].loudness;
743 /* get higher loudness to consider */
745 nb_pow = LRA_HIGHER_PRC * nb_powers / 100. + 0.5;
746 for (i = HIST_SIZE - 1; i >= 0; i--) {
747 n -= ebur128->i3000.histogram[i].count;
749 ebur128->lra_high = ebur128->i3000.histogram[i].loudness;
754 // XXX: show low & high on the graph?
755 ebur128->loudness_range = ebur128->lra_high - ebur128->lra_low;
759 /* dual-mono correction */
760 if (nb_channels == 1 && ebur128->dual_mono) {
761 loudness_400 -= ebur128->pan_law;
762 loudness_3000 -= ebur128->pan_law;
765 #define LOG_FMT "TARGET:%d LUFS M:%6.1f S:%6.1f I:%6.1f %s LRA:%6.1f LU"
767 /* push one video frame */
768 if (ebur128->do_video) {
773 int y_loudness_lu_graph, y_loudness_lu_gauge;
775 if (ebur128->gauge_type == GAUGE_TYPE_MOMENTARY) {
776 gauge_value = loudness_400 - ebur128->target;
778 gauge_value = loudness_3000 - ebur128->target;
781 y_loudness_lu_graph = lu_to_y(ebur128, loudness_3000 - ebur128->target);
782 y_loudness_lu_gauge = lu_to_y(ebur128, gauge_value);
784 /* draw the graph using the short-term loudness */
785 p = pic->data[0] + ebur128->graph.y*pic->linesize[0] + ebur128->graph.x*3;
786 for (y = 0; y < ebur128->graph.h; y++) {
787 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_graph, y);
789 memmove(p, p + 3, (ebur128->graph.w - 1) * 3);
790 memcpy(p + (ebur128->graph.w - 1) * 3, c, 3);
791 p += pic->linesize[0];
794 /* draw the gauge using either momentary or short-term loudness */
795 p = pic->data[0] + ebur128->gauge.y*pic->linesize[0] + ebur128->gauge.x*3;
796 for (y = 0; y < ebur128->gauge.h; y++) {
797 const uint8_t *c = get_graph_color(ebur128, y_loudness_lu_gauge, y);
799 for (x = 0; x < ebur128->gauge.w; x++)
800 memcpy(p + x*3, c, 3);
801 p += pic->linesize[0];
804 /* draw textual info */
805 if (ebur128->scale == SCALE_TYPE_ABSOLUTE) {
806 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
807 LOG_FMT " ", // padding to erase trailing characters
808 ebur128->target, loudness_400, loudness_3000,
809 ebur128->integrated_loudness, "LUFS", ebur128->loudness_range);
811 drawtext(pic, PAD, PAD - PAD/2, FONT16, font_colors,
812 LOG_FMT " ", // padding to erase trailing characters
813 ebur128->target, loudness_400-ebur128->target, loudness_3000-ebur128->target,
814 ebur128->integrated_loudness-ebur128->target, "LU", ebur128->loudness_range);
817 /* set pts and push frame */
819 clone = av_frame_clone(pic);
821 return AVERROR(ENOMEM);
822 ret = ff_filter_frame(outlink, clone);
827 if (ebur128->metadata) { /* happens only once per filter_frame call */
829 #define META_PREFIX "lavfi.r128."
831 #define SET_META(name, var) do { \
832 snprintf(metabuf, sizeof(metabuf), "%.3f", var); \
833 av_dict_set(&insamples->metadata, name, metabuf, 0); \
836 #define SET_META_PEAK(name, ptype) do { \
837 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
839 for (ch = 0; ch < nb_channels; ch++) { \
840 snprintf(key, sizeof(key), \
841 META_PREFIX AV_STRINGIFY(name) "_peaks_ch%d", ch); \
842 SET_META(key, ebur128->name##_peaks[ch]); \
847 SET_META(META_PREFIX "M", loudness_400);
848 SET_META(META_PREFIX "S", loudness_3000);
849 SET_META(META_PREFIX "I", ebur128->integrated_loudness);
850 SET_META(META_PREFIX "LRA", ebur128->loudness_range);
851 SET_META(META_PREFIX "LRA.low", ebur128->lra_low);
852 SET_META(META_PREFIX "LRA.high", ebur128->lra_high);
854 SET_META_PEAK(sample, SAMPLES);
855 SET_META_PEAK(true, TRUE);
858 if (ebur128->scale == SCALE_TYPE_ABSOLUTE) {
859 av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
860 av_ts2timestr(pts, &outlink->time_base),
861 ebur128->target, loudness_400, loudness_3000,
862 ebur128->integrated_loudness, "LUFS", ebur128->loudness_range);
864 av_log(ctx, ebur128->loglevel, "t: %-10s " LOG_FMT,
865 av_ts2timestr(pts, &outlink->time_base),
866 ebur128->target, loudness_400-ebur128->target, loudness_3000-ebur128->target,
867 ebur128->integrated_loudness-ebur128->target, "LU", ebur128->loudness_range);
870 #define PRINT_PEAKS(str, sp, ptype) do { \
871 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
872 av_log(ctx, ebur128->loglevel, " " str ":"); \
873 for (ch = 0; ch < nb_channels; ch++) \
874 av_log(ctx, ebur128->loglevel, " %5.1f", DBFS(sp[ch])); \
875 av_log(ctx, ebur128->loglevel, " dBFS"); \
879 PRINT_PEAKS("SPK", ebur128->sample_peaks, SAMPLES);
880 PRINT_PEAKS("FTPK", ebur128->true_peaks_per_frame, TRUE);
881 PRINT_PEAKS("TPK", ebur128->true_peaks, TRUE);
882 av_log(ctx, ebur128->loglevel, "\n");
886 return ff_filter_frame(ctx->outputs[ebur128->do_video], insamples);
889 static int query_formats(AVFilterContext *ctx)
891 EBUR128Context *ebur128 = ctx->priv;
892 AVFilterFormats *formats;
893 AVFilterChannelLayouts *layouts;
894 AVFilterLink *inlink = ctx->inputs[0];
895 AVFilterLink *outlink = ctx->outputs[0];
898 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_NONE };
899 static const int input_srate[] = {48000, -1}; // ITU-R BS.1770 provides coeff only for 48kHz
900 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
902 /* set optional output video format */
903 if (ebur128->do_video) {
904 formats = ff_make_format_list(pix_fmts);
905 if ((ret = ff_formats_ref(formats, &outlink->incfg.formats)) < 0)
907 outlink = ctx->outputs[1];
910 /* set input and output audio formats
911 * Note: ff_set_common_* functions are not used because they affect all the
912 * links, and thus break the video format negotiation */
913 formats = ff_make_format_list(sample_fmts);
914 if ((ret = ff_formats_ref(formats, &inlink->outcfg.formats)) < 0 ||
915 (ret = ff_formats_ref(formats, &outlink->incfg.formats)) < 0)
918 layouts = ff_all_channel_layouts();
919 if ((ret = ff_channel_layouts_ref(layouts, &inlink->outcfg.channel_layouts)) < 0 ||
920 (ret = ff_channel_layouts_ref(layouts, &outlink->incfg.channel_layouts)) < 0)
923 formats = ff_make_format_list(input_srate);
924 if ((ret = ff_formats_ref(formats, &inlink->outcfg.samplerates)) < 0 ||
925 (ret = ff_formats_ref(formats, &outlink->incfg.samplerates)) < 0)
931 static av_cold void uninit(AVFilterContext *ctx)
934 EBUR128Context *ebur128 = ctx->priv;
936 /* dual-mono correction */
937 if (ebur128->nb_channels == 1 && ebur128->dual_mono) {
938 ebur128->i400.rel_threshold -= ebur128->pan_law;
939 ebur128->i3000.rel_threshold -= ebur128->pan_law;
940 ebur128->lra_low -= ebur128->pan_law;
941 ebur128->lra_high -= ebur128->pan_law;
944 av_log(ctx, AV_LOG_INFO, "Summary:\n\n"
945 " Integrated loudness:\n"
947 " Threshold: %5.1f LUFS\n\n"
950 " Threshold: %5.1f LUFS\n"
951 " LRA low: %5.1f LUFS\n"
952 " LRA high: %5.1f LUFS",
953 ebur128->integrated_loudness, ebur128->i400.rel_threshold,
954 ebur128->loudness_range, ebur128->i3000.rel_threshold,
955 ebur128->lra_low, ebur128->lra_high);
957 #define PRINT_PEAK_SUMMARY(str, sp, ptype) do { \
961 if (ebur128->peak_mode & PEAK_MODE_ ## ptype ## _PEAKS) { \
962 for (ch = 0; ch < ebur128->nb_channels; ch++) \
963 maxpeak = FFMAX(maxpeak, sp[ch]); \
964 av_log(ctx, AV_LOG_INFO, "\n\n " str " peak:\n" \
965 " Peak: %5.1f dBFS", \
970 PRINT_PEAK_SUMMARY("Sample", ebur128->sample_peaks, SAMPLES);
971 PRINT_PEAK_SUMMARY("True", ebur128->true_peaks, TRUE);
972 av_log(ctx, AV_LOG_INFO, "\n");
974 av_freep(&ebur128->y_line_ref);
975 av_freep(&ebur128->ch_weighting);
976 av_freep(&ebur128->true_peaks);
977 av_freep(&ebur128->sample_peaks);
978 av_freep(&ebur128->true_peaks_per_frame);
979 av_freep(&ebur128->i400.histogram);
980 av_freep(&ebur128->i3000.histogram);
981 for (i = 0; i < ebur128->nb_channels; i++) {
982 av_freep(&ebur128->i400.cache[i]);
983 av_freep(&ebur128->i3000.cache[i]);
985 av_frame_free(&ebur128->outpicref);
986 #if CONFIG_SWRESAMPLE
987 av_freep(&ebur128->swr_buf);
988 swr_free(&ebur128->swr_ctx);
992 static const AVFilterPad ebur128_inputs[] = {
995 .type = AVMEDIA_TYPE_AUDIO,
996 .filter_frame = filter_frame,
997 .config_props = config_audio_input,
1002 const AVFilter ff_af_ebur128 = {
1004 .description = NULL_IF_CONFIG_SMALL("EBU R128 scanner."),
1005 .priv_size = sizeof(EBUR128Context),
1008 .query_formats = query_formats,
1009 .inputs = ebur128_inputs,
1011 .priv_class = &ebur128_class,
1012 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,